def convertToMultiFit(i3data, x_size, y_size, frame, nm_per_pixel, inverted=False):
"""
Create a 3D-DAOSTORM, sCMOS or Spliner analysis compatible peak array from I3 data.
Notes:
(1) This uses the non-drift corrected positions.
(2) This sets the initial fitting error to zero and the status to RUNNING.
"""
i3data = maskData(i3data, (i3data['fr'] == frame))
peaks = numpy.zeros((i3data.size, utilC.getNPeakPar()))
peaks[:,utilC.getBackgroundIndex()] = i3data['bg']
peaks[:,utilC.getHeightIndex()] = i3data['h']
peaks[:,utilC.getZCenterIndex()] = i3data['z'] * 0.001
if inverted:
peaks[:,utilC.getXCenterIndex()] = y_size - i3data['x']
peaks[:,utilC.getYCenterIndex()] = x_size - i3data['y']
ax = i3data['ax']
ww = i3data['w']
peaks[:,utilC.getYWidthIndex()] = 0.5*numpy.sqrt(ww*ww/ax)/nm_per_pixel
peaks[:,utilC.getXWidthIndex()] = 0.5*numpy.sqrt(ww*ww*ax)/nm_per_pixel
else:
peaks[:,utilC.getYCenterIndex()] = i3data['x'] - 1
peaks[:,utilC.getXCenterIndex()] = i3data['y'] - 1
ax = i3data['ax']
ww = i3data['w']
peaks[:,utilC.getXWidthIndex()] = 0.5*numpy.sqrt(ww*ww/ax)/nm_per_pixel
peaks[:,utilC.getYWidthIndex()] = 0.5*numpy.sqrt(ww*ww*ax)/nm_per_pixel
return peaks
def createFromMultiFit(molecules,
x_size,
y_size,
frame,
nm_per_pixel,
inverted=False):
"""
Create an I3 data from the output of 3D-DAOSTORM, sCMOS or Spliner.
"""
n_molecules = molecules.shape[0]
h = molecules[:, 0]
if inverted:
xc = y_size - molecules[:, utilC.getXCenterIndex()]
yc = x_size - molecules[:, utilC.getYCenterIndex()]
wx = 2.0 * molecules[:, utilC.getXWidthIndex()] * nm_per_pixel
wy = 2.0 * molecules[:, utilC.getYWidthIndex()] * nm_per_pixel
else:
xc = molecules[:, utilC.getYCenterIndex()] + 1
yc = molecules[:, utilC.getXCenterIndex()] + 1
wx = 2.0 * molecules[:, utilC.getYWidthIndex()] * nm_per_pixel
wy = 2.0 * molecules[:, utilC.getXWidthIndex()] * nm_per_pixel
bg = molecules[:, utilC.getBackgroundIndex()]
zc = molecules[:, utilC.getZCenterIndex(
)] * 1000.0 # fitting is done in um, insight works in nm
st = numpy.round(molecules[:, utilC.getStatusIndex()])
err = molecules[:, utilC.getErrorIndex()]
#
# Calculate peak area, which is saved in the "a" field.
#
# Note that this is assuming that the peak is a 2D gaussian. This
# will not calculate the correct area for a Spline..
#
parea = 2.0 * 3.14159 * h * molecules[:, utilC.getXWidthIndex(
)] * molecules[:, utilC.getYWidthIndex()]
ax = wy / wx
ww = numpy.sqrt(wx * wy)
i3data = createDefaultI3Data(xc.size)
posSet(i3data, 'x', xc)
posSet(i3data, 'y', yc)
posSet(i3data, 'z', zc)
setI3Field(i3data, 'h', h)
setI3Field(i3data, 'bg', bg)
setI3Field(i3data, 'fi', st)
setI3Field(i3data, 'a', parea)
setI3Field(i3data, 'w', ww)
setI3Field(i3data, 'ax', ax)
setI3Field(i3data, 'fr', frame)
setI3Field(i3data, 'i', err)
return i3data
def createFromMultiFit(molecules, x_size, y_size, frame, nm_per_pixel, inverted=False):
"""
Create an I3 data from the output of 3D-DAOSTORM, sCMOS or Spliner.
"""
n_molecules = molecules.shape[0]
h = molecules[:,0]
if inverted:
xc = y_size - molecules[:,utilC.getXCenterIndex()]
yc = x_size - molecules[:,utilC.getYCenterIndex()]
wx = 2.0*molecules[:,utilC.getXWidthIndex()]*nm_per_pixel
wy = 2.0*molecules[:,utilC.getYWidthIndex()]*nm_per_pixel
else:
xc = molecules[:,utilC.getYCenterIndex()] + 1
yc = molecules[:,utilC.getXCenterIndex()] + 1
wx = 2.0*molecules[:,utilC.getYWidthIndex()]*nm_per_pixel
wy = 2.0*molecules[:,utilC.getXWidthIndex()]*nm_per_pixel
bg = molecules[:,utilC.getBackgroundIndex()]
zc = molecules[:,utilC.getZCenterIndex()] * 1000.0 # fitting is done in um, insight works in nm
st = numpy.round(molecules[:,utilC.getStatusIndex()])
err = molecules[:,utilC.getErrorIndex()]
#
# Calculate peak area, which is saved in the "a" field.
#
# Note that this is assuming that the peak is a 2D gaussian. This
# will not calculate the correct area for a Spline..
#
parea = 2.0*3.14159*h*molecules[:,utilC.getXWidthIndex()]*molecules[:,utilC.getYWidthIndex()]
ax = wy/wx
ww = numpy.sqrt(wx*wy)
i3data = createDefaultI3Data(xc.size)
posSet(i3data, 'x', xc)
posSet(i3data, 'y', yc)
posSet(i3data, 'z', zc)
setI3Field(i3data, 'h', h)
setI3Field(i3data, 'bg', bg)
setI3Field(i3data, 'fi', st)
setI3Field(i3data, 'a', parea)
setI3Field(i3data, 'w', ww)
setI3Field(i3data, 'ax', ax)
setI3Field(i3data, 'fr', frame)
setI3Field(i3data, 'i', err)
return i3data
def __init__(self, parameters):
fitting.PeakFinder.__init__(self, parameters)
self.height_rescale = []
self.mfilter = []
self.mfilter_z = []
self.z_value = []
# Load the spline.
self.s_to_psf = splineToPSF.loadSpline(parameters.getAttr("spline"))
# Update margin based on the spline size.
old_margin = self.margin
self.margin = int((self.s_to_psf.getSize() + 1)/4 + 2)
self.mfilter_z = parameters.getAttr("z_value", [0.0])
for zval in self.mfilter_z:
self.z_value.append(self.s_to_psf.getScaledZ(zval))
if parameters.hasAttr("peak_locations"):
# Correct for any difference in the margins.
self.peak_locations[:,utilC.getXCenterIndex()] += self.margin - old_margin
self.peak_locations[:,utilC.getYCenterIndex()] += self.margin - old_margin
# Provide the "correct" starting z value.
self.peak_locations[:,utilC.getZCenterIndex()] = self.z_value[0]
def __init__(self, parameters):
fitting.PeakFinder.__init__(self, parameters)
self.height_rescale = []
self.mfilter = []
self.mfilter_z = []
self.z_value = []
# Load the spline.
self.s_to_psf = splineToPSF.loadSpline(parameters.getAttr("spline"))
# Update margin based on the spline size.
old_margin = self.margin
self.margin = int((self.s_to_psf.getSize() + 1)/4 + 2)
self.mfilter_z = parameters.getAttr("z_value", [0.0])
for zval in self.mfilter_z:
self.z_value.append(self.s_to_psf.getScaledZ(zval))
if parameters.hasAttr("peak_locations"):
# Correct for any difference in the margins.
self.peak_locations[:,utilC.getXCenterIndex()] += self.margin - old_margin
self.peak_locations[:,utilC.getYCenterIndex()] += self.margin - old_margin
# Provide the "correct" starting z value.
self.peak_locations[:,utilC.getZCenterIndex()] = self.z_value[0]
def analyzeImage(self,
new_image,
bg_estimate=None,
save_residual=False,
verbose=False):
"""
image - The image to analyze.
bg_estimate - (Optional) An estimate of the background.
save_residual - (Optional) Save the residual image after peak fitting, default is False.
return - [Found peaks, Image residual]
"""
#
# Pad out arrays so that we can better analyze localizations
# near the edge of the original image.
#
image = padArray(new_image, self.margin)
residual = padArray(new_image, self.margin)
if bg_estimate is not None:
bg_estimate = padArray(bg_estimate, self.margin)
self.peak_finder.newImage(image)
self.peak_fitter.newImage(image)
if save_residual:
resid_dax = daxwriter.DaxWriter("residual.dax", residual.shape[0],
residual.shape[1])
peaks = False
for i in range(self.peak_finder.iterations):
if save_residual:
resid_dax.addFrame(residual)
no_bg_image = self.peak_finder.subtractBackground(
residual, bg_estimate)
[found_new_peaks,
peaks] = self.peak_finder.findPeaks(no_bg_image, peaks)
if isinstance(peaks, numpy.ndarray):
[peaks, residual] = self.peak_fitter.fitPeaks(peaks)
if verbose:
if isinstance(peaks, numpy.ndarray):
print(" peaks:", i, found_new_peaks, peaks.shape[0])
else:
print(" peaks:", i, found_new_peaks, "NA")
if not found_new_peaks:
break
if save_residual:
resid_dax.addFrame(residual)
resid_dax.close()
if isinstance(peaks, numpy.ndarray):
peaks[:, utilC.getXCenterIndex()] -= float(self.margin)
peaks[:, utilC.getYCenterIndex()] -= float(self.margin)
return [peaks, residual]
def __init__(self, parameters):
# Initialized from parameters.
self.find_max_radius = parameters.getAttr("find_max_radius", 5) # Radius (in pixels) over which the maxima is maximal.
self.iterations = parameters.getAttr("iterations") # Maximum number of cycles of peak finding, fitting and subtraction to perform.
self.sigma = parameters.getAttr("sigma") # Peak sigma (in pixels).
self.threshold = parameters.getAttr("threshold") # Peak minimum threshold (height, in camera units).
self.z_value = parameters.getAttr("z_value", 0.0) # The starting z value to use for peak fitting.
# Other member variables.
self.background = None # Current estimate of the image background.
self.image = None # The original image.
self.margin = PeakFinderFitter.margin # Size of the unanalyzed "edge" around the image.
self.neighborhood = PeakFinder.unconverged_dist * self.sigma # Radius for marking neighbors as unconverged.
self.new_peak_radius = PeakFinder.new_peak_dist # Minimum allowed distance between new peaks and current peaks.
self.parameters = parameters # Keep access to the parameters object.
self.peak_locations = None # Initial peak locations, as explained below.
self.peak_mask = None # Mask for limiting peak identification to a particular AOI.
self.taken = None # Spots in the image where a peak has already been added.
#
# This is for is you already know where your want fitting to happen, as
# for example in a bead calibration movie and you just want to use the
# approximate locations as inputs for fitting.
#
# peak_locations is a text file with the peak x, y, height and background
# values as white spaced columns (x and y positions are in pixels as
# determined using visualizer).
#
# 1.0 2.0 1000.0 100.0
# 10.0 5.0 2000.0 200.0
# ...
#
if parameters.hasAttr("peak_locations"):
print("Using peak starting locations specified in", parameters.getAttr("peak_locations"))
# Only do one cycle of peak finding as we'll always return the same locations.
if (self.iterations != 1):
print("WARNING: setting number of iterations to 1!")
self.iterations = 1
# Load peak x,y locations.
peak_locs = numpy.loadtxt(parameters.getAttr("peak_locations"), ndmin = 2)
print(peak_locs.shape)
# Create peak array.
self.peak_locations = numpy.zeros((peak_locs.shape[0],
utilC.getNPeakPar()))
self.peak_locations[:,utilC.getXCenterIndex()] = peak_locs[:,1] + self.margin
self.peak_locations[:,utilC.getYCenterIndex()] = peak_locs[:,0] + self.margin
self.peak_locations[:,utilC.getHeightIndex()] = peak_locs[:,2]
self.peak_locations[:,utilC.getBackgroundIndex()] = peak_locs[:,3]
self.peak_locations[:,utilC.getXWidthIndex()] = numpy.ones(peak_locs.shape[0]) * self.sigma
self.peak_locations[:,utilC.getYWidthIndex()] = numpy.ones(peak_locs.shape[0]) * self.sigma
def convertToMultiFit(i3data,
x_size,
y_size,
frame,
nm_per_pixel,
inverted=False):
"""
Create a 3D-DAOSTORM, sCMOS or Spliner analysis compatible peak array from I3 data.
Notes:
(1) This uses the non-drift corrected positions.
(2) This sets the initial fitting error to zero and the status to RUNNING.
"""
i3data = maskData(i3data, (i3data['fr'] == frame))
peaks = numpy.zeros((i3data.size, utilC.getNPeakPar()))
peaks[:, utilC.getBackgroundIndex()] = i3data['bg']
peaks[:, utilC.getHeightIndex()] = i3data['h']
peaks[:, utilC.getZCenterIndex()] = i3data['z'] * 0.001
if inverted:
peaks[:, utilC.getXCenterIndex()] = y_size - i3data['x']
peaks[:, utilC.getYCenterIndex()] = x_size - i3data['y']
ax = i3data['ax']
ww = i3data['w']
peaks[:, utilC.getYWidthIndex()] = 0.5 * numpy.sqrt(
ww * ww / ax) / nm_per_pixel
peaks[:, utilC.getXWidthIndex()] = 0.5 * numpy.sqrt(
ww * ww * ax) / nm_per_pixel
else:
peaks[:, utilC.getYCenterIndex()] = i3data['x'] - 1
peaks[:, utilC.getXCenterIndex()] = i3data['y'] - 1
ax = i3data['ax']
ww = i3data['w']
peaks[:, utilC.getXWidthIndex()] = 0.5 * numpy.sqrt(
ww * ww / ax) / nm_per_pixel
peaks[:, utilC.getYWidthIndex()] = 0.5 * numpy.sqrt(
ww * ww * ax) / nm_per_pixel
return peaks
def findPeaks(self, peaks):
# Adjust x,y for margin.
peaks[:, utilC.getXCenterIndex()] += float(self.margin)
peaks[:, utilC.getYCenterIndex()] += float(self.margin)
# Convert z (in nano-meters) to spline z units.
zi = utilC.getZCenterIndex()
peaks[:, zi] = self.s_to_psfs[0].getScaledZ(peaks[:, zi])
# Split peaks into per-channel peaks.
return self.mpu.splitPeaks(peaks)
def analyzeImage(self,
new_image,
bg_estimate=None,
save_residual=False,
verbose=False):
image = fitting.padArray(new_image, self.peak_finder.margin)
if bg_estimate is not None:
bg_estimate = fitting.padArray(bg_estimate,
self.peak_finder.margin)
self.peak_finder.newImage(image, bg_estimate)
self.peak_fitter.newImage(image)
#
# This is a lot simpler than 3D-DAOSTORM as we only do one pass,
# hopefully the compressed sensing (FISTA) deconvolution finds all the
# peaks and then we do a single pass of fitting.
#
if True:
peaks = self.peak_finder.findPeaks()
[fit_peaks, residual] = self.peak_fitter.fitPeaks(peaks)
#
# This is for testing if just using FISTA followed by the center
# of mass calculation is basically as good as also doing the
# additional MLE spline fitting step.
#
# The short answer is that it appears that it is not. It about
# 1.3x worse in XY and about 4x worse in Z.
#
else:
fit_peaks = self.peak_finder.findPeaks()
# Adjust z scale.
z_index = utilC.getZCenterIndex()
z_size = (self.peak_fitter.spline.shape[2] - 1.0)
status_index = utilC.getStatusIndex()
fit_peaks[:, z_index] = z_size * fit_peaks[:, z_index]
# Mark as converged.
fit_peaks[:, status_index] = 1.0
residual = None
#
# Subtract margin so that peaks are in the right
# place with respect to the original image.
#
fit_peaks[:, utilC.getXCenterIndex()] -= float(self.peak_finder.margin)
fit_peaks[:, utilC.getYCenterIndex()] -= float(self.peak_finder.margin)
return [fit_peaks, residual]
def analyzeImage(self, movie_reader, save_residual=False, verbose=False):
"""
Analyze an "image" and return a list of the found localizations.
Internally the list of localizations is a multiple of the number of planes.
"""
#
# Load and scale the images.
#
[images, fit_peaks_images] = self.loadImages(movie_reader)
#
# Load and scale the background estimates.
#
bg_estimates = self.loadBackgroundEstimates(movie_reader)
self.peak_finder.newImages(images)
self.peak_fitter.newImages(images)
peaks = False
for i in range(self.peak_finder.iterations):
if verbose:
print(" iteration", i)
# Update background estimate.
for j in range(self.n_planes):
self.peak_finder.subtractBackground(
images[j] - fit_peaks_images[j], bg_estimates[j], j)
# Find new peaks.
[found_new_peaks,
peaks] = self.peak_finder.findPeaks(fit_peaks_images, peaks)
# Fit new peaks.
if isinstance(peaks, numpy.ndarray):
if verbose:
print(" found", peaks.shape[0] / self.n_planes)
[peaks, fit_peaks_images] = self.peak_fitter.fitPeaks(peaks)
if verbose:
print(" fit", peaks.shape[0] / self.n_planes)
if not found_new_peaks:
break
if isinstance(peaks, numpy.ndarray):
peaks[:, utilC.getXCenterIndex()] -= float(self.margin)
peaks[:, utilC.getYCenterIndex()] -= float(self.margin)
#
# sa_utilities.std_analysis doesn't do anything with the second
# argument, historically the residual, so just return None.
#
return [peaks, None]
def analyzeImage(self, new_image, bg_estimate = None, save_residual = False, verbose = False):
image = fitting.padArray(new_image, self.peak_finder.margin)
if bg_estimate is not None:
bg_estimate = fitting.padArray(bg_estimate, self.peak_finder.margin)
self.peak_finder.newImage(image, bg_estimate)
self.peak_fitter.newImage(image)
#
# This is a lot simpler than 3D-DAOSTORM as we only do one pass,
# hopefully the compressed sensing (FISTA) deconvolution finds all the
# peaks and then we do a single pass of fitting.
#
if True:
peaks = self.peak_finder.findPeaks()
[fit_peaks, residual] = self.peak_fitter.fitPeaks(peaks)
#
# This is for testing if just using FISTA followed by the center
# of mass calculation is basically as good as also doing the
# additional MLE spline fitting step.
#
# The short answer is that it appears that it is not. It about
# 1.3x worse in XY and about 4x worse in Z.
#
else:
fit_peaks = self.peak_finder.findPeaks()
# Adjust z scale.
z_index = utilC.getZCenterIndex()
z_size = (self.peak_fitter.spline.shape[2] - 1.0)
status_index = utilC.getStatusIndex()
fit_peaks[:,z_index] = z_size*fit_peaks[:,z_index]
# Mark as converged.
fit_peaks[:,status_index] = 1.0
residual = None
#
# Subtract margin so that peaks are in the right
# place with respect to the original image.
#
fit_peaks[:,utilC.getXCenterIndex()] -= float(self.peak_finder.margin)
fit_peaks[:,utilC.getYCenterIndex()] -= float(self.peak_finder.margin)
return [fit_peaks, residual]
def getPeaks(self, threshold, margin):
"""
Extract peaks from the deconvolved image and create
an array that can be used by a peak fitter.
FIXME: Need to compensate for up-sampling parameter in x,y.
"""
fx = self.getXVector()
# Get area, position, height.
fd_peaks = fdUtil.getPeaks(fx, threshold, margin)
num_peaks = fd_peaks.shape[0]
peaks = numpy.zeros((num_peaks, utilC.getNPeakPar()))
peaks[:, utilC.getXWidthIndex()] = numpy.ones(num_peaks)
peaks[:, utilC.getYWidthIndex()] = numpy.ones(num_peaks)
peaks[:, utilC.getXCenterIndex()] = fd_peaks[:, 2]
peaks[:, utilC.getYCenterIndex()] = fd_peaks[:, 1]
# Calculate height.
#
# FIXME: Typically the starting value for the peak height will be
# under-estimated unless a large enough number of FISTA
# iterations is performed to completely de-convolve the image.
#
h_index = utilC.getHeightIndex()
#peaks[:,h_index] = fd_peaks[:,0]
for i in range(num_peaks):
peaks[i, h_index] = fd_peaks[i, 0] * self.psf_heights[int(
round(fd_peaks[i, 3]))]
# Calculate z (0.0 - 1.0).
if (fx.shape[2] > 1):
peaks[:, utilC.getZCenterIndex()] = fd_peaks[:, 3] / (
float(fx.shape[2]) - 1.0)
# Background term calculation.
bg_index = utilC.getBackgroundIndex()
for i in range(num_peaks):
ix = int(round(fd_peaks[i, 1]))
iy = int(round(fd_peaks[i, 2]))
peaks[i, bg_index] = self.background[ix, iy]
return peaks
def analyzeImage(self, new_image, bg_estimate = None, save_residual = False, verbose = False):
#
# Pad out arrays so that we can better analyze localizations
# near the edge of the original image.
#
image = padArray(new_image, self.margin)
residual = padArray(new_image, self.margin)
if bg_estimate is not None:
bg_estimate = padArray(bg_estimate, self.margin)
self.peak_finder.newImage(image)
self.peak_fitter.newImage(image)
if save_residual:
resid_dax = daxwriter.DaxWriter("residual.dax",
residual.shape[0],
residual.shape[1])
peaks = False
for i in range(self.peak_finder.iterations):
if save_residual:
resid_dax.addFrame(residual)
no_bg_image = self.peak_finder.subtractBackground(residual, bg_estimate)
[found_new_peaks, peaks] = self.peak_finder.findPeaks(no_bg_image, peaks)
if isinstance(peaks, numpy.ndarray):
[peaks, residual] = self.peak_fitter.fitPeaks(peaks)
if verbose:
if isinstance(peaks, numpy.ndarray):
print(" peaks:", i, found_new_peaks, peaks.shape[0])
else:
print(" peaks:", i, found_new_peaks, "NA")
if not found_new_peaks:
break
if save_residual:
resid_dax.addFrame(residual)
resid_dax.close()
if isinstance(peaks, numpy.ndarray):
peaks[:,utilC.getXCenterIndex()] -= float(self.margin)
peaks[:,utilC.getYCenterIndex()] -= float(self.margin)
return [peaks, residual]
def analyze(base_name, mlist_name, settings_name):
parameters = params.ParametersMultiplane().initFromFile(settings_name)
# Set to only analyze 1 frame, 1 iteration of peak finding.
parameters.setAttr("max_frame", "int", 1)
parameters.setAttr("iterations", "int", 1)
# Analyze one frame.
finder = PFPeakFinder(parameters)
fitter = PFPeakFitter(parameters)
finder_fitter = findPeaksStd.MPFinderFitter(parameters, finder, fitter)
reader = findPeaksStd.MPMovieReader(base_name=base_name,
parameters=parameters)
stdAnalysis.standardAnalysis(finder_fitter, reader, mlist_name, parameters)
# Evaluate found vs fit parameters.
hi = utilC.getHeightIndex()
xi = utilC.getXCenterIndex()
yi = utilC.getYCenterIndex()
# First identify peak pairs.
fi_x = fitted_peaks[:, xi]
fi_y = fitted_peaks[:, yi]
fo_x = found_peaks[:, xi]
fo_y = found_peaks[:, yi]
dd = utilC.peakToPeakDist(fo_x, fo_y, fi_x, fi_y)
di = utilC.peakToPeakIndex(fo_x, fo_y, fi_x, fi_y)
print(numpy.mean(dd), fi_x.size, fo_x.size)
fo_h = []
fi_h = []
for i in range(dd.size):
if (dd[i] < 2.0):
fo_h.append(found_peaks[i, hi])
fi_h.append(fitted_peaks[di[i], hi])
fi_h = numpy.array(fi_h)
fo_h = numpy.array(fo_h)
print(numpy.mean(fi_h), fi_h.size)
print(numpy.mean(fo_h), fo_h.size)
print(numpy.mean(fi_h) / numpy.mean(fo_h))
def getPeaks(self, threshold, margin):
fx = self.getXVector()
# Get area, position, height.
fd_peaks = fdUtil.getPeaks(fx, threshold, margin)
num_peaks = fd_peaks.shape[0]
peaks = numpy.zeros((num_peaks, utilC.getNResultsPar()))
peaks[:, utilC.getXWidthIndex()] = numpy.ones(num_peaks)
peaks[:, utilC.getYWidthIndex()] = numpy.ones(num_peaks)
peaks[:, utilC.getXCenterIndex()] = fd_peaks[:, 2]
peaks[:, utilC.getYCenterIndex()] = fd_peaks[:, 1]
# Calculate height.
#
# FIXME: Typically the starting value for the peak height will be
# under-estimated unless a large enough number of FISTA
# iterations is performed to completely de-convolve the image.
#
h_index = utilC.getHeightIndex()
#peaks[:,h_index] = fd_peaks[:,0]
for i in range(num_peaks):
peaks[i, h_index] = fd_peaks[i, 0] * self.psf_heights[int(
round(fd_peaks[i, 3]))]
# Calculate z (0.0 - 1.0).
peaks[:,
utilC.getZCenterIndex()] = fd_peaks[:, 3] / (float(fx.shape[2]) -
1.0)
# Background term calculation.
bg_index = utilC.getBackgroundIndex()
for i in range(num_peaks):
ix = int(round(fd_peaks[i, 1]))
iy = int(round(fd_peaks[i, 2]))
peaks[i, bg_index] = self.background[ix, iy]
return peaks
def analyzeImage(self, movie_reader, save_residual=False, verbose=False):
# Load images & background estimates.
[images, fit_peaks_images] = self.loadImages(movie_reader)
# Pass images to fitter.
self.peak_fitter.newImages(images)
# Load ground truth localization positions.
truth_peaks = movie_reader.getLocalizations()
# Adjust to work with multi-plane.
fit_peaks = self.peak_finder.findPeaks(truth_peaks.copy())
# Calculate best fits peaks.
[fit_peaks, fit_peaks_images] = self.peak_fitter.fitPeaks(fit_peaks)
# Adjust for margin.
fit_peaks[:, utilC.getXCenterIndex()] -= float(self.margin)
fit_peaks[:, utilC.getYCenterIndex()] -= float(self.margin)
return [fit_peaks, None]
def __init__(self, parameters):
super(MPPeakFinder, self).__init__(parameters)
self.atrans = [None]
self.backgrounds = []
self.bg_filter = None
self.bg_filter_sigma = parameters.getAttr("bg_filter_sigma")
self.check_mode = False
self.height_rescale = []
self.images = []
self.mapping_filename = None
self.mfilters = []
self.mfilters_z = []
self.mpu = None
self.n_channels = 0
self.s_to_psfs = []
self.variances = []
self.vfilters = []
self.xt = []
self.yt = []
self.z_values = []
# Print warning about check mode
if self.check_mode:
print("Warning: Running in check mode.")
# Load the splines.
for spline_attr in mpUtilC.getSplineAttrs(parameters):
self.s_to_psfs.append(
splineToPSF.loadSpline(parameters.getAttr(spline_attr)))
self.n_channels += 1
# Assert that all the splines are the same size.
for i in range(1, len(self.s_to_psfs)):
assert (self.s_to_psfs[0].getSize() == self.s_to_psfs[i].getSize())
#
# Update margin based on the spline size. Note the assumption
# that all of the splines are the same size, or at least smaller
# than the spline for plane 0.
#
old_margin = self.margin
self.margin = int((self.s_to_psfs[0].getSize() + 1) / 4 + 2)
# Load the plane to plane mapping data & create affine transform objects.
mappings = {}
if parameters.hasAttr("mapping"):
if os.path.exists(parameters.getAttr("mapping")):
self.mapping_filename = parameters.getAttr("mapping")
with open(parameters.getAttr("mapping"), 'rb') as fp:
mappings = pickle.load(fp)
# Use self.margin - 1, because we added 1 to the x,y coordinates when we saved them.
for i in range(self.n_channels - 1):
self.xt.append(
mpUtilC.marginCorrect(mappings["0_" + str(i + 1) + "_x"],
self.margin - 1))
self.yt.append(
mpUtilC.marginCorrect(mappings["0_" + str(i + 1) + "_y"],
self.margin - 1))
self.atrans.append(
affineTransformC.AffineTransform(xt=self.xt[i], yt=self.yt[i]))
#
# Note the assumption that the splines for each plane all use
# the same z scale / have the same z range.
#
self.mfilters_z = parameters.getAttr("z_value", [0.0])
for zval in self.mfilters_z:
self.z_values.append(self.s_to_psfs[0].getScaledZ(zval))
if parameters.hasAttr("peak_locations"):
# Correct for any difference in the margins.
self.peak_locations[:, utilC.getXCenterIndex(
)] += self.margin - old_margin
self.peak_locations[:, utilC.getYCenterIndex(
)] += self.margin - old_margin
# Provide the "correct" starting z value.
#
# FIXME: Should also allow the user to specify the peak z location
# as any fixed value could be so far off for some of the
# localizations that the fitting will fail.
#
self.peak_locations[:, utilC.getZCenterIndex()] = self.z_values[0]
def measurePSF(movie_name, zfile_name, movie_mlist, psf_name, want2d = False, aoi_size = 12, z_range = 750.0, z_step = 50.0):
"""
The actual z range is 2x z_range (i.e. from -z_range to z_range).
"""
# Load dax file, z offset file and molecule list file.
dax_data = datareader.inferReader(movie_name)
z_offsets = None
if os.path.exists(zfile_name):
try:
z_offsets = numpy.loadtxt(zfile_name, ndmin = 2)[:,1]
except IndexError:
z_offsets = None
print("z offsets were not loaded.")
i3_data = readinsight3.loadI3File(movie_mlist)
if want2d:
print("Measuring 2D PSF")
else:
print("Measuring 3D PSF")
#
# Go through the frames identifying good peaks and adding them
# to the average psf. For 3D molecule z positions are rounded to
# the nearest 50nm.
#
z_mid = int(z_range/z_step)
max_z = 2 * z_mid + 1
average_psf = numpy.zeros((max_z,4*aoi_size,4*aoi_size))
peaks_used = 0
totals = numpy.zeros(max_z)
[dax_x, dax_y, dax_l] = dax_data.filmSize()
for curf in range(dax_l):
# Select localizations in current frame & not near the edges.
mask = (i3_data['fr'] == curf+1) & (i3_data['x'] > aoi_size) & (i3_data['x'] < (dax_x - aoi_size - 1)) & (i3_data['y'] > aoi_size) & (i3_data['y'] < (dax_y - aoi_size - 1))
xr = i3_data['x'][mask]
yr = i3_data['y'][mask]
# Use the z offset file if it was specified, otherwise use localization z positions.
if z_offsets is None:
if (curf == 0):
print("Using fit z locations.")
zr = i3_data['z'][mask]
else:
if (curf == 0):
print("Using z offset file.")
zr = numpy.ones(xr.size) * z_offsets[curf]
ht = i3_data['h'][mask]
# Remove localizations that are too close to each other.
in_peaks = numpy.zeros((xr.size,util_c.getNPeakPar()))
in_peaks[:,util_c.getXCenterIndex()] = xr
in_peaks[:,util_c.getYCenterIndex()] = yr
in_peaks[:,util_c.getZCenterIndex()] = zr
in_peaks[:,util_c.getHeightIndex()] = ht
out_peaks = util_c.removeNeighbors(in_peaks, 2*aoi_size)
#out_peaks = util_c.removeNeighbors(in_peaks, aoi_size)
print(curf, "peaks in", in_peaks.shape[0], ", peaks out", out_peaks.shape[0])
# Use remaining localizations to calculate spline.
image = dax_data.loadAFrame(curf).astype(numpy.float64)
xr = out_peaks[:,util_c.getXCenterIndex()]
yr = out_peaks[:,util_c.getYCenterIndex()]
zr = out_peaks[:,util_c.getZCenterIndex()]
ht = out_peaks[:,util_c.getHeightIndex()]
for i in range(xr.size):
xf = xr[i]
yf = yr[i]
zf = zr[i]
xi = int(xf)
yi = int(yf)
if want2d:
zi = 0
else:
zi = int(round(zf/z_step) + z_mid)
# check the z is in range
if (zi > -1) and (zi < max_z):
# get localization image
mat = image[xi-aoi_size:xi+aoi_size,
yi-aoi_size:yi+aoi_size]
# zoom in by 2x
psf = scipy.ndimage.interpolation.zoom(mat, 2.0)
# re-center image
psf = scipy.ndimage.interpolation.shift(psf, (-2.0*(xf-xi), -2.0*(yf-yi)), mode='nearest')
# add to average psf accumulator
average_psf[zi,:,:] += psf
totals[zi] += 1
# Force PSF to be zero (on average) at the boundaries.
for i in range(max_z):
edge = numpy.concatenate((average_psf[i,0,:],
average_psf[i,-1,:],
average_psf[i,:,0],
average_psf[i,:,-1]))
average_psf[i,:,:] -= numpy.mean(edge)
# Normalize the PSF.
if want2d:
max_z = 1
for i in range(max_z):
print(i, totals[i])
if (totals[i] > 0.0):
average_psf[i,:,:] = average_psf[i,:,:]/numpy.sum(numpy.abs(average_psf[i,:,:]))
average_psf = average_psf/numpy.max(average_psf)
# Save PSF (in image form).
if True:
import storm_analysis.sa_library.daxwriter as daxwriter
dxw = daxwriter.DaxWriter(os.path.join(os.path.dirname(psf_name), "psf.dax"),
average_psf.shape[1],
average_psf.shape[2])
for i in range(max_z):
dxw.addFrame(1000.0 * average_psf[i,:,:] + 100)
dxw.close()
# Save PSF.
if want2d:
psf_dict = {"psf" : average_psf[0,:,:],
"type" : "2D"}
else:
cur_z = -z_range
z_vals = []
for i in range(max_z):
z_vals.append(cur_z)
cur_z += z_step
psf_dict = {"psf" : average_psf,
"pixel_size" : 0.080, # 1/2 the camera pixel size in nm.
"type" : "3D",
"zmin" : -z_range,
"zmax" : z_range,
"zvals" : z_vals}
pickle.dump(psf_dict, open(psf_name, 'wb'))
def psfLocalizations(i3_filename, mapping_filename, frame = 1, aoi_size = 8, movie_filename = None):
# Load localizations.
i3_reader = readinsight3.I3Reader(i3_filename)
# Load mapping.
mappings = {}
if os.path.exists(mapping_filename):
with open(mapping_filename, 'rb') as fp:
mappings = pickle.load(fp)
else:
print("Mapping file not found, single channel data?")
# Try and determine movie frame size.
i3_metadata = readinsight3.loadI3Metadata(i3_filename)
if i3_metadata is None:
if movie_filename is None:
raise Exception("I3 metadata not found and movie filename is not specified.")
else:
movie_fp = datareader.inferReader(movie_filename)
[movie_y, movie_x] = movie_fp.filmSize()[:2]
else:
movie_data = i3_metadata.find("movie")
# FIXME: These may be transposed?
movie_x = int(movie_data.find("movie_x").text)
movie_y = int(movie_data.find("movie_y").text)
# Load localizations in the requested frame.
locs = i3_reader.getMoleculesInFrame(frame)
print("Loaded", locs.size, "localizations.")
# Remove localizations that are too close to each other.
in_locs = numpy.zeros((locs["x"].size, util_c.getNPeakPar()))
in_locs[:,util_c.getXCenterIndex()] = locs["x"]
in_locs[:,util_c.getYCenterIndex()] = locs["y"]
out_locs = util_c.removeNeighbors(in_locs, 2 * aoi_size)
xf = out_locs[:,util_c.getXCenterIndex()]
yf = out_locs[:,util_c.getYCenterIndex()]
#
# Remove localizations that are too close to the edge or
# outside of the image in any of the channels.
#
is_good = numpy.ones(xf.size, dtype = numpy.bool)
for i in range(xf.size):
# Check in Channel 0.
if (xf[i] < aoi_size) or (xf[i] + aoi_size >= movie_x):
is_good[i] = False
continue
if (yf[i] < aoi_size) or (yf[i] + aoi_size >= movie_y):
is_good[i] = False
continue
# Check other channels.
for key in mappings:
if not is_good[i]:
break
coeffs = mappings[key]
[ch1, ch2, axis] = key.split("_")
if (ch1 == "0"):
if (axis == "x"):
xm = coeffs[0] + coeffs[1]*xf[i] + coeffs[2]*yf[i]
if (xm < aoi_size) or (xm + aoi_size >= movie_x):
is_good[i] = False
break
elif (axis == "y"):
ym = coeffs[0] + coeffs[1]*xf[i] + coeffs[2]*yf[i]
if (ym < aoi_size) or (ym + aoi_size >= movie_y):
is_good[i] = False
break
#
# Save localizations for each channel.
#
gx = xf[is_good]
gy = yf[is_good]
basename = os.path.splitext(i3_filename)[0]
with writeinsight3.I3Writer(basename + "_c1_psf.bin") as w3:
w3.addMoleculesWithXY(gx, gy)
index = 1
while ("0_" + str(index) + "_x" in mappings):
cx = mappings["0_" + str(index) + "_x"]
cy = mappings["0_" + str(index) + "_y"]
#cx = mappings[str(index) + "_0" + "_x"]
#cy = mappings[str(index) + "_0" + "_y"]
xm = cx[0] + cx[1] * gx + cx[2] * gy
ym = cy[0] + cy[1] * gx + cy[2] * gy
with writeinsight3.I3Writer(basename + "_c" + str(index+1) + "_psf.bin") as w3:
w3.addMoleculesWithXY(xm, ym)
index += 1
#
# Print localizations that were kept.
#
print(gx.size, "localizations were kept:")
for i in range(gx.size):
print("ch0: {0:.2f} {1:.2f}".format(gx[i], gy[i]))
index = 1
while ("0_" + str(index) + "_x" in mappings):
cx = mappings["0_" + str(index) + "_x"]
cy = mappings["0_" + str(index) + "_y"]
xm = cx[0] + cx[1] * gx[i] + cx[2] * gy[i]
ym = cy[0] + cy[1] * gx[i] + cy[2] * gy[i]
print("ch" + str(index) + ": {0:.2f} {1:.2f}".format(xm, ym))
index += 1
print("")
print("")
def __init__(self, parameters):
# Member variables.
self.background = None # Current estimate of the image background.
self.image = None # The original image.
self.iterations = parameters.iterations # Maximum number of cycles of peak finding, fitting and subtraction to perform.
self.margin = PeakFinderFitter.margin # Size of the unanalyzed "edge" around the image.
self.neighborhood = PeakFinder.unconverged_dist * parameters.sigma # Radius for marking neighbors as unconverged.
self.new_peak_radius = PeakFinder.new_peak_dist # Minimum allowed distance between new peaks and current peaks.
self.parameters = parameters # Keep access to the parameters object.
self.peak_locations = None # Initial peak locations, as explained below.
self.peak_mask = None # Mask for limiting peak identification to a particular AOI.
self.sigma = parameters.sigma # Peak sigma (in pixels).
self.taken = None # Spots in the image where a peak has already been added.
self.threshold = parameters.threshold # Peak minimum threshold (height, in camera units).
self.z_value = 0.0 # The starting z value to use for peak fitting.
# Radius (in pixels) over which the maxima is maximal.
if hasattr(parameters, "find_max_radius"):
self.find_max_radius = float(parameters.find_max_radius)
else:
self.find_max_radius = 5
#
# This is for is you already know where your want fitting to happen, as
# for example in a bead calibration movie and you just want to use the
# approximate locations as inputs for fitting.
#
# peak_locations is a text file with the peak x, y, height and background
# values as white spaced columns (x and y positions are in pixels as
# determined using visualizer).
#
# 1.0 2.0 1000.0 100.0
# 10.0 5.0 2000.0 200.0
# ...
#
if hasattr(parameters, "peak_locations"):
print("Using peak starting locations specified in",
parameters.peak_locations)
# Only do one cycle of peak finding as we'll always return the same locations.
if (self.iterations != 1):
print("WARNING: setting number of iterations to 1!")
self.iterations = 1
# Load peak x,y locations.
peak_locs = numpy.loadtxt(parameters.peak_locations, ndmin=2)
print(peak_locs.shape)
# Create peak array.
self.peak_locations = numpy.zeros(
(peak_locs.shape[0], util_c.getNResultsPar()))
self.peak_locations[:, util_c.getXCenterIndex(
)] = peak_locs[:, 1] + self.margin
self.peak_locations[:, util_c.getYCenterIndex(
)] = peak_locs[:, 0] + self.margin
self.peak_locations[:, util_c.getHeightIndex()] = peak_locs[:, 2]
self.peak_locations[:, util_c.getBackgroundIndex()] = peak_locs[:,
3]
self.peak_locations[:, util_c.getXWidthIndex()] = numpy.ones(
peak_locs.shape[0]) * self.sigma
self.peak_locations[:, util_c.getYWidthIndex()] = numpy.ones(
peak_locs.shape[0]) * self.sigma
def measurePSF(movie_name,
zfile_name,
movie_mlist,
psf_name,
want2d=False,
aoi_size=12,
z_range=750.0,
z_step=50.0):
"""
The actual z range is 2x z_range (i.e. from -z_range to z_range).
"""
# Load dax file, z offset file and molecule list file.
dax_data = datareader.inferReader(movie_name)
z_offsets = None
if os.path.exists(zfile_name):
try:
z_offsets = numpy.loadtxt(zfile_name, ndmin=2)[:, 1]
except IndexError:
z_offsets = None
print("z offsets were not loaded.")
i3_data = readinsight3.loadI3File(movie_mlist)
if want2d:
print("Measuring 2D PSF")
else:
print("Measuring 3D PSF")
#
# Go through the frames identifying good peaks and adding them
# to the average psf. For 3D molecule z positions are rounded to
# the nearest 50nm.
#
z_mid = int(z_range / z_step)
max_z = 2 * z_mid + 1
average_psf = numpy.zeros((max_z, 4 * aoi_size, 4 * aoi_size))
peaks_used = 0
totals = numpy.zeros(max_z)
[dax_x, dax_y, dax_l] = dax_data.filmSize()
for curf in range(dax_l):
# Select localizations in current frame & not near the edges.
mask = (i3_data['fr'] == curf + 1) & (i3_data['x'] > aoi_size) & (
i3_data['x'] <
(dax_x - aoi_size - 1)) & (i3_data['y'] >
aoi_size) & (i3_data['y'] <
(dax_y - aoi_size - 1))
xr = i3_data['x'][mask]
yr = i3_data['y'][mask]
# Use the z offset file if it was specified, otherwise use localization z positions.
if z_offsets is None:
if (curf == 0):
print("Using fit z locations.")
zr = i3_data['z'][mask]
else:
if (curf == 0):
print("Using z offset file.")
zr = numpy.ones(xr.size) * z_offsets[curf]
ht = i3_data['h'][mask]
# Remove localizations that are too close to each other.
in_peaks = numpy.zeros((xr.size, util_c.getNPeakPar()))
in_peaks[:, util_c.getXCenterIndex()] = xr
in_peaks[:, util_c.getYCenterIndex()] = yr
in_peaks[:, util_c.getZCenterIndex()] = zr
in_peaks[:, util_c.getHeightIndex()] = ht
out_peaks = util_c.removeNeighbors(in_peaks, 2 * aoi_size)
#out_peaks = util_c.removeNeighbors(in_peaks, aoi_size)
print(curf, "peaks in", in_peaks.shape[0], ", peaks out",
out_peaks.shape[0])
# Use remaining localizations to calculate spline.
image = dax_data.loadAFrame(curf).astype(numpy.float64)
xr = out_peaks[:, util_c.getXCenterIndex()]
yr = out_peaks[:, util_c.getYCenterIndex()]
zr = out_peaks[:, util_c.getZCenterIndex()]
ht = out_peaks[:, util_c.getHeightIndex()]
for i in range(xr.size):
xf = xr[i]
yf = yr[i]
zf = zr[i]
xi = int(xf)
yi = int(yf)
if want2d:
zi = 0
else:
zi = int(round(zf / z_step) + z_mid)
# check the z is in range
if (zi > -1) and (zi < max_z):
# get localization image
mat = image[xi - aoi_size:xi + aoi_size,
yi - aoi_size:yi + aoi_size]
# zoom in by 2x
psf = scipy.ndimage.interpolation.zoom(mat, 2.0)
# re-center image
psf = scipy.ndimage.interpolation.shift(
psf, (-2.0 * (xf - xi), -2.0 * (yf - yi)), mode='nearest')
# add to average psf accumulator
average_psf[zi, :, :] += psf
totals[zi] += 1
# Force PSF to be zero (on average) at the boundaries.
for i in range(max_z):
edge = numpy.concatenate((average_psf[i, 0, :], average_psf[i, -1, :],
average_psf[i, :, 0], average_psf[i, :, -1]))
average_psf[i, :, :] -= numpy.mean(edge)
# Normalize the PSF.
if want2d:
max_z = 1
for i in range(max_z):
print(i, totals[i])
if (totals[i] > 0.0):
average_psf[i, :, :] = average_psf[i, :, :] / numpy.sum(
numpy.abs(average_psf[i, :, :]))
average_psf = average_psf / numpy.max(average_psf)
# Save PSF (in image form).
if True:
import storm_analysis.sa_library.daxwriter as daxwriter
dxw = daxwriter.DaxWriter(
os.path.join(os.path.dirname(psf_name), "psf.dax"),
average_psf.shape[1], average_psf.shape[2])
for i in range(max_z):
dxw.addFrame(1000.0 * average_psf[i, :, :] + 100)
dxw.close()
# Save PSF.
if want2d:
psf_dict = {"psf": average_psf[0, :, :], "type": "2D"}
else:
cur_z = -z_range
z_vals = []
for i in range(max_z):
z_vals.append(cur_z)
cur_z += z_step
psf_dict = {
"psf": average_psf,
"pixel_size": 0.080, # 1/2 the camera pixel size in nm.
"type": "3D",
"zmin": -z_range,
"zmax": z_range,
"zvals": z_vals
}
with open(psf_name, 'wb') as fp:
pickle.dump(psf_dict, fp)
peaks_used = 0
total = 0.0
[dax_x, dax_y, dax_l] = dax_data.filmSize()
while (curf < dax_l) and (peaks_used < min_peaks):
# Select localizations in current frame & not near the edges.
mask = (i3_data['fr'] == curf) & (i3_data['x'] > aoi_size) & (
i3_data['x'] < (dax_y - aoi_size - 1)) & (i3_data['y'] > aoi_size) & (
i3_data['y'] < (dax_x - aoi_size - 1))
xr = i3_data['x'][mask]
yr = i3_data['y'][mask]
ht = i3_data['h'][mask]
# Remove localizations that are too close to each other.
in_peaks = numpy.zeros((xr.size, util_c.getNResultsPar()))
in_peaks[:, util_c.getXCenterIndex()] = xr
in_peaks[:, util_c.getYCenterIndex()] = yr
in_peaks[:, util_c.getHeightIndex()] = ht
out_peaks = util_c.removeNeighbors(in_peaks, aoi_size)
print(curf, in_peaks.shape, out_peaks.shape)
# Use remaining localizations to calculate spline.
image = dax_data.loadAFrame(curf - 1).astype(numpy.float64)
xr = out_peaks[:, util_c.getXCenterIndex()]
yr = out_peaks[:, util_c.getYCenterIndex()]
ht = out_peaks[:, util_c.getHeightIndex()]
for i in range(xr.size):
average_psf = numpy.zeros((4*aoi_size,4*aoi_size))
curf = 1
peaks_used = 0
total = 0.0
[dax_x, dax_y, dax_l] = dax_data.filmSize()
while (curf < dax_l) and (peaks_used < min_peaks):
# Select localizations in current frame & not near the edges.
mask = (i3_data['fr'] == curf) & (i3_data['x'] > aoi_size) & (i3_data['x'] < (dax_y - aoi_size - 1)) & (i3_data['y'] > aoi_size) & (i3_data['y'] < (dax_x - aoi_size - 1))
xr = i3_data['x'][mask]
yr = i3_data['y'][mask]
ht = i3_data['h'][mask]
# Remove localizations that are too close to each other.
in_peaks = numpy.zeros((xr.size,util_c.getNResultsPar()))
in_peaks[:,util_c.getXCenterIndex()] = xr
in_peaks[:,util_c.getYCenterIndex()] = yr
in_peaks[:,util_c.getHeightIndex()] = ht
out_peaks = util_c.removeNeighbors(in_peaks, aoi_size)
print(curf, in_peaks.shape, out_peaks.shape)
# Use remaining localizations to calculate spline.
image = dax_data.loadAFrame(curf-1).astype(numpy.float64)
xr = out_peaks[:,util_c.getXCenterIndex()]
yr = out_peaks[:,util_c.getYCenterIndex()]
ht = out_peaks[:,util_c.getHeightIndex()]
for i in range(xr.size):